Performance Testing for GDB

Size: px

Start display at page:

Download "Performance Testing for GDB"

Stuart Bond
8 years ago
Views:

1 Performance Testing for GDB Yao Qi CodeSourcery/MentorGraphics

2 1 Overview Introduction Goals 2 3

3 Introduction Goals Introduction We need something to test and measure GDB perf first, No standard mechanism to show the performance, source:accurate-measurement.jpg Proposed the perf testing framework in Aug and Sep 213, Write some perf test cases and patches for perf improvements from Sep to Nov. if you cannot measure it, you cannot improve it. Lord Kelvin Thank Doug and Sanimir for the comments and patient review.

jpg Proposed the perf testing framework in Aug and Sep 213, Write some perf test cases and patches for

4 Goals Overview Introduction Goals Know the new perf testing framework in GDB and how to write perf test cases on top of it, Demonstrate how do they show the perf improvements, CPUT time (s) original patched Number of single steps Show how do we use it to track GDB performance

show the perf improvements, CPUT time (s) 4 3 3 2 2 1 original patched 1 1

5 Overview easy to write test case for a certain aspect, i.e. symbol lookup or single step basic measurements are provided (such as cpu time and memory usage), but test specific measurement can be added too. both native debugging and remote debugging are supported, test case has warm up optionally, save results in different formats, the executable in test case can be pre-compiled, source:shelter1.jpg

6 Use DejaGNU to invoke compiler and start GDB (and/or GDBserver), GDB loads a python script in which some operations are performed, The framework collects the performance data and save them in the desired format, Detect performance regressions, compile test case gdb.perf/foo.c start gdb and gdbserver load program run gdb.perf/foo.py lib/gdb.exp lib/perftest.exp gdb.perf/foo.exp

in the desired format, Detect performance regressions, compile test case gdb.perf/foo.

7 : skip-prologue.exp 1 PerfTest::assemble { 2 if { [ gdb_compile " $srcdir / $subdir / $srcfile " ${ binfile } executable { debug }]!= "" } { 3 return -1 4 } COMPILE 6 return 7 } { 8 clean_restart $binfile START UP 9 1 if![ runto_main ] { 11 fail "Can t run to main " 12 return } 14 } { 1 global SKIP_PROLOGUE_COUNT 16 RUN 17 set test " run " 18 gdb_test_multiple " python SkipPrologue $ $SKIP_PROLOGUE_COUNT $.run () " $test { 19 -re " Breakpoint $decimal at \[^\ n\] *\n" { 2 # GDB prints some messages on breakpoint creation. 21 # Consume the output to avoid internal buffer full. 22 exp_continue 23 } 24 -re ".*$gdb_prompt $" { 2 pass $test 26 } 27 } 28 }

$[ runto_main ] { 11 fail "Can t run to main " 12 return -1 13 } 14 } { 1 global SKIP_PROLOGUE_COUNT 16 RUN 17 set test " run " 18 gdb_test_multiple " python$

8 : skip-prologue.py 1 from perftest import perftest 2 3 class SkipPrologue ( perftest. TestCaseWithBasicMeasurements): 4 def init (self, count ): super( SkipPrologue, self ). init ("skip - prologue ") 6 self. count = count 7 Extend from 8 def _test ( self ): TestCaseWithBasicMeasurements 9 for _ in range (1, self. count ): 1 # Insert breakpoints on function f1 and f2. 11 bp1 = gdb. Breakpoint ("f1 ") 12 bp2 = gdb. Breakpoint ("f2 ") 13 # Remove them. 14 bp1. delete () 1 bp2. delete () Test def warm_up ( self ): 18 self. _test () 19 Measure the test 2 def execute_test ( self ): 21 for i in range (1, 4): 22 gdb. execute ("set code - cache off ") 23 gdb. execute ("set code - cache on ") 24 self. measure. measure ( self._test, i) 1 TestCase name 9 measure exuecte testcase () run () 8 warm up () 7 6 TestCaseWithBasicMeasurements SkipPrologue 1 warm up () execute test ()

11 bp1 = gdb. Breakpoint ("f1 ") 12 bp2 = gdb. Breakpoint ("f2 ") 13 # Remove them. 14 bp1. delete () 1 bp2. delete () Test 16 17 def warm_up ( self ): 18 self.

9 Cache code access for skip-prologue Before After read some bytes RSP target read some bytes read target dcache RSP target parse the instruction parse the instruction 6 Original Patched yes is it part of prologue? yes is it part of prologue? no stop no stop CPU time (s) note that only x86 and x86 64 make use that SKIP_PROLOGUE_COUNT

Original Patched yes is it part of prologue?

10 Cache code access for disassemble 9 8 Original Patched 3. 3 Original Patched Number of m packets CPU time (s) evaluate_subexp_standardhandle_inferior_event c_parse_internal evaluate_subexp_standard handle_inferior_event c_parse_internal

11 My workflow of improving perf Examine the RSP log and locate some unnecessary packets (gdb) si Sending packet: $qtstatus#49... Sending packet: $Z,8483c7,1#b4...Packet received: OK Sending packet: $Z,4ceb6b,1#6e...Packet received: OK Sending packet: $QPassSignals:... Sending packet: $vcont;s:p1b1.1b1;c#2... Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $mbfffef4,4#c... Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $z,8483c7,1#d4...packet received: OK Sending packet: $z,4ceb6b,1#8e...packet received: OK Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Investigate and hack the code, Show the perf improved by the patch, CPUT time (s) original patched Typical profling doesn t help % gdbserver libc so [.] strcmp_sse4_2 strcmp_sse4_2-1.92% gdbserver libc so [.] vfprintf - vfprintf % _IO_vsprintf 2.6% handle_tracepoint_query % x % _IO_vsnprintf Number of single steps

..packet received: E1 Sending packet: $mbfffef4,4#c... Sending packet: $qxfer:traceframe-info:read::,fff#b...packet received: E1 Sending packet: $qxfer:traceframe-info:read::,fff#b.

12 Read multiple cache lines in dcache xfer memory Before After 6 Original Patched 3 Original Patched 3 Number of m packets CPU Time (s) Cache line size Cache line size

4 3 2 1 CPU Time (s) 2 2 1 1 16 32 64 128 26 12 124 248

13 Track and compare GDB performance Still have no way to track and compare perf data A web application to monitor and analyze the performance, Written by python and easy to set up, It doesn t meet our needs perfectly, and it is not actively maintained,

the performance, Written by python and easy to set up, It

14 Observations and future work observations Command thread doesn t scale, Switch thread by command thread N CPU Time (s) Actual dcache is conservative for multi-threaded code, 1 enum target_xfer_status 2 dcache_read_memory_partial ( struct target_ops *ops, DCACHE * dcache, 3 CORE_ADDR memaddr, gdb_byte * myaddr, 4 ULONGEST len, ULONGEST * xfered_len ) { 6 /* If this is a different inferior from what we ve recorded, 7 flush the cache. */ 8 9 (! ptid_equal ( inferior_ptid if, dcache -> ptid )) 1 { 11 dcache_invalidate ( dcache ); 12 dcache -> ptid = inferior_ptid ; Remove? 13 } Number of threads 3 2 Current(remote) Patched(remote) Disassemble across threads 1 static int 2 do_captured_thread_select ( struct ui_out * uiout, void * tidstr ) 3 { 4... /* Since the current thread may have changed, see if there is any 6 exited thread we can now delete. */ 7 prune_threads (); 8 9 return GDB_RC_OK ; 1 } CPU Time (s) Number of threads future work Compare perf data and alarm if regression is found. More perf test cases

$ptid_equal ( inferior_ptid if, dcache -> ptid )) 1 { 11 dcache_invalidate ( dcache ); 12 dcache -> ptid = inferior_ptid ; Remove?$

15 Framework Overview

Real-time Debugging using GDB Tracepoints and other Eclipse features

Real-time Debugging using GDB Tracepoints and other Eclipse features GCC Summit 2010 2010-010-26 marc.khouzam@ericsson.com Summary Introduction Advanced debugging features Non-stop multi-threaded debugging